Stoa, a Helsinki company that builds AI software for construction project information, announced a €3.5 million pre-seed round (about $3.9 million) on November 16, 2026, led by Lifeline Ventures. The software reads a project’s drawings, specifications, contracts, schedule and site records and answers questions from them, citing the document and page behind each answer. That is the lookup a project engineer or site manager runs many times a day when a foreman asks which detail governs a wall or which spec section covers a penetration. Stoa says the platform is in use on more than 20 sites across the Nordics, including data center projects.
What it does
Stoa connects to the document systems a project already runs and syncs from them read-only, keeping the permissions each system already set. The demo on its homepage shows M-Files, Autodesk Construction Cloud, SharePoint, Congrid for quality and safety records, and Primavera P6 for the master schedule, and the page also names Dalux. Stoa says its own team connects the systems and a project goes live two working days after the agreement.
A user asks a question in plain language and gets an answer in which each sentence cites a document and page. The homepage example asks what fire rating the level 2 partitions need. The answer cites a floor plan, the partition spec and a wall-type sheet.
Clicking a wall type or door tag on a drawing opens the detail or clause that governs it beside the plan. A change feed lists reissued drawings and revised specs, filtered to the user’s scope, and shows two revisions side by side. Stoa labels each change summary in that feed as read off the document by a model and warns that it can be wrong.
The software also drafts. Prompts in the demo include an RFI response, a contract review, a weekly site report, a risk analysis of the programme and a draft claim. On its data center page, Stoa says the software flags where the record supports a claim and assembles the claim from the drawings, contract and schedule.
Evidence so far
The evidence is Stoa’s press release and website. The performance figures come from what Stoa calls interviews with design partners in July 2026, and the portfolio figure from its current customer list. Customers shown on the site include Ailecon, A-Insinöörit, CapMan, Gaga Muller and The nooa Company. Pat Nicholson, Nordics Director at GagaMuller, says in the release that the tool became routine on site quickly because it works through the portals the team already used.
These figures come from Stoa:
| Figure | Value | What Stoa says it measures |
|---|---|---|
| Time saved per project engineer | 2 to 8 hours a week | Hours back per person, most of it time spent finding where something was written down |
| Problem raised to decision made | 40% shorter | Faster decisions because the drawings and clause arrive with the question |
| Mistakes caught before site | More than €3 million | Conflicts flagged between the programme and the documents it depends on |
| Onboarding | 5 times faster | New staff get answers by asking the project in plain language |
| Project value on the platform | More than €4 billion | Stoa's current portfolio, including Nordic data center sites |
The 2 to 8 hours is a per-person weekly range, and Stoa says most of it is time engineers used to spend finding where something was written down. At the low end that is 5% of a 40-hour week, and at the high end 20%. The 40% applies to the stretch from a problem being raised to a decision, a span that can include waiting on a designer or an owner. The €4 billion measures the size of the jobs on the platform and carries no information about time saved.
Interview figures record what users estimate they save. The number of people interviewed, and how Stoa timed the 40% and five-times figures, are not yet public.
Co-founder and Chief Product Officer Taavi Paananen described the problem the product targets in the release: “It’s just in the wrong system at the wrong time.” He was talking about the information that would prevent site mistakes. Tracy Young and Ralph Gootee, co-founders of PlanGrid, invested as angels. (Autodesk bought PlanGrid for $875 million in 2018, the release notes.)
How to measure it on your projects
The figure to track is question turnaround: the calendar hours from a field question being raised to an answer that cites the governing drawing, spec section or contract clause. Questions that go to the design team as formal RFIs belong in a separate log, using the method in measuring RFI turnaround before an AI pilot.
| Metric | Definition | Baseline to capture before the pilot |
|---|---|---|
| Question turnaround | Calendar hours from a field question being raised to an answer that cites the governing drawing or spec section or contract clause | Two weeks of logged questions on one active job |
| Search minutes per question | Minutes the project engineer spent finding the answer in the document systems | Same two-week log |
| Questions that became RFIs | Share of logged questions the team could not settle from the documents and sent to the design team | Same two-week log |
| Answer error rate | Share of answers a reviewer later found pointed to the wrong or superseded document | Spot-check 10 logged answers against the current document set |
Search minutes per question is the closest match to Stoa’s hours claim, and the answer error rate covers the risk Stoa itself flags in the change feed. An answer drawn from a superseded sheet can send a crew to build to the wrong revision, so check a sample of answers against the current set.
For two weeks on one active job, have the project engineer log every question that comes in, with the time it arrived, the time it was answered, the minutes spent searching and whether it became an RFI. Repeat the log during the pilot on the same job.
For tools that work on the RFI and submittal side of the same documents, see the posts on Shackleton and Ageiro’s ARK Watcher.
Availability and cost
Stoa prices by quote after a demo on the customer’s own project documents. It says data is stored and processed in EU regions, never used to train any model, and deleted within 30 days of a customer’s request. All of the sites Stoa describes are in the Nordics. The company says it will use the round to expand internationally, with the Nordics, the UK and the US as its focus markets. Stoa’s reasoning is that the technology companies building data centers operate globally, so one customer relationship can open several markets.